Dear aspiring developer,

You have the mental chops to do development: you create Excel sheets, write SQL, etc. So it is disappointing to watch developers ship real software with AI while your best is a pretty web app. AI can mimic their years of training, but that still isn’t enough: developers also have tools to build software, a machine to build it on, and prebuilt code to build from. dk gives you all three. It’s time to build real software.

Jonah Beckford

Coming soon

dk Prompt Studio. Describe what you want to build and get a paste-ready “mini-plan” for your AI agent.

How it works

  1. Onboard with a GitHub account
  2. Your AI agent writes dk scripts
  3. A remote GitHub Actions machine builds it
  4. Ship real software

Compared to traditional development

You Traditional software developer
Onboarding IT gives you GitHub Enterprise account, or you use Student/Free account IT gives them corporate machine + GitHub Enterprise account
Code Creator AI agent Themselves and/or AI agent
Hardware GitHub Actions Beefy, always-on development machine
Operating System Windows + maybe macOS Linux + Windows + maybe macOS
Security Sandboxing GitHub Actions Docker and/or Virtual Machines
Coding Style High Throughput: many agents at once. Code must be self-instrumenting (ie. logs, metrics, detailed errors) to cut debug cycles Low Latency: they code one or two things simultaneously with fast, local feedback

Why dk?

Commands

get-object Fetch a prebuilt object (a tool, library, or binary) from the registry.
remote Build on a remote GitHub Actions machine instead of your own.
add Add a GitHub release verified with a SLSA-2 attestation to your workspace.
restore Rebuild only what changed: fast, incremental builds.
run-object Run a built object locally or remote.
distribute Package a build into a signed, checksummed distribution.

Full command reference →